Activity Based Interfaces in Online Social Networks
by
Jawad Laraqui
Submitted to the Department of Electrical Engineering and Computer Science
in Partial Fulfillment of the Requirements for the Degree of
Master of Engineering in Electrical Engineering and Computer Science
at the Massachusetts Institute of Technology
May 2007
©2007 Massachusetts Institute of Technology
All rights reserved.
Author____________________________________________________________
Department of Electrical Engineering and Computer Science
May 24, 2007
Certified by____________________________________________________________
Judith Donath
Assistant Professor of Media Arts and Sciences
Thesis Supervisor
Accepted by____________________________________________________________
Arthur C. Smith
Professor of Electrical Engineering
Chairman, Department Committee on Graduate Theses
2
Activity Based Interfaces in Online Social Networks
by
Jawad Laraqui
Submitted to the
Department of Electrical Engineering and Computer Science
May 24, 2007
In Partial Fulfillment of the Requirements for the Degree of
Master of Engineering in Electrical Engineering and Computer Science
ABSTRACT
The goal of the project is to explore how activity-based interfaces can create more
meaningful experiences for the users and builders of online social networking sites.
Medina, a social-networking site based on the idea of exchanging knowledge, explores
new interfaces for visualizing connections between people and ideas. The site constantly
measures interactions between people and their interests in order to create a more
accurate picture of what relationships and information are important.
Thesis Supervisor: Judith Donath
Title: Assistant Professor of Media Arts and Science
3
4
Acknowledgements
Thank you to the following people:
Judith Donath for her guidance, focus, and patience. I am truly privileged to have
had the opportunity to learn from her.
Drew Harry, Aaron Zinman, Orkan Telhan, and Dietmar Offenhuber for all their
help and friendship.
Meg for her patience and love.
My family for their eternal support and love.
5
Contents
1.
2.
Introduction.............................................................................................................. 7
Background .............................................................................................................. 8
2.1. Definition of Social Ties.................................................................................... 8
2.2. Signaling Theory ............................................................................................. 12
3. Signaling in Social Networks.................................................................................. 14
3.1. User Level Signaling ....................................................................................... 14
3.2. Service Level Signaling ................................................................................... 15
3.3. Site Navigation Interfaces................................................................................ 17
4. Sconex: signaling from a builder’s perspective ....................................................... 20
4.1. History ............................................................................................................ 20
4.2. Money Shapes Design ..................................................................................... 25
4.3. Stimulating Information Flow and Visibility.................................................... 27
4.3.1. Increasing Number of users .......................................................................... 27
4.3.2. Increasing Page Views Per Person ................................................................ 29
4.3.3. New User Experience ................................................................................... 30
4.4. Challenges from Design Decisions .................................................................. 31
4.5. Signal Theory Analysis.................................................................................... 33
5. Medina: exploring activity based interfaces ............................................................ 36
5.1. A marketplace for knowledge .......................................................................... 36
5.2. Activity-Based Interface .................................................................................. 39
5.3. Site Functionality............................................................................................. 41
5.3.1. Platform Specifications................................................................................. 41
5.3.2. Feature Set.................................................................................................... 41
5.3.3. Measuring Activity ....................................................................................... 44
5.4. Activity Visualization ...................................................................................... 46
5.5. Medina Evaluation........................................................................................... 49
6. Conclusion ............................................................................................................. 50
7. Bibliography .......................................................................................................... 51
6
1. Introduction
Online social networks are not a new phenomenon. They have evolved slowly into
their current form as the popularity of the Internet increased, and as technical advances
allowed services to support more complicated feature sets. In the 90s, the fledgling online
communities were either easy to maintain message boards or static “homepage” style web
sites hosted by extremely large companies like Geocities or Lycos. These large
companies had the financial resources required to run these services. As technology
became more inexpensive it made hosting more advanced social networks with more
complex feature sets financially feasible for small companies. Indeed, over the last
decade, innovation in this space has always come from small startups. The social
networking behemoths like MySpace and Facebook share the same humble beginnings as
smaller communities like LiveJournal or Xanga. The final incarnation of all of these
services has been the result of a long journey where challenges included creating a user
base from scratch, marketing on extremely tight budgets, and facing fierce competition.
However, when online social networks are discussed only half the story is told.
Interface design is analyzed from the user’s perspective and fails to take into account the
external constraints that shape the service. There are two realms in which design tradeoffs
are evaluated. From the user’s perspective, the purpose of a social networking service is
to enable self-expression. At the site builder’s level, the service is a tool used to turn a
profit, to attract advertisers and investors, and to acquire new members. In this realm,
providing a compelling user experience is just one of many constraints. Thus it can be
expected that many of the design problems found in online social networking sites arise
7
from the disconnect and inconsistencies between these incentives. In fact, the users are
only peripherally involved in the major decisions that shape the world they operate in.
The power struggle is analogous to a dictatorship where the needs of the population are
subjected to the choices of the powerful. Additionally, it is hard for the site builders
themselves to gain enough perspective to accurately pinpoint high level problems during
the design process. The rush to build up a user base before money runs out often leaves
many awkward design choices in the system.
We will explore the possibility of democratizing these design struggles by using
activity-based interfaces in online social networking sites. Our analysis will begin by
revisiting the definitions of social ties in light of this new dual view of constraints, and by
introducing Signaling Theory as a framework for analyzing such services. Then, we will
formalize the needs of users and site builders using Signaling Theory. Next, we will take
a detailed look at Sconex, a current social networking service, to understand the evolving
constraints a service must face and the far reaching effects of interface design. Finally,
we will discuss the implementation of an activity-based service called Medina and
evaluate its effectiveness in aligning the needs of site builders and users.
2. Background
2.1. Definition of Social Ties
Traditionally, social networking sites have used the same underlying network
structure: a graph where the nodes are the users, and the connections are the explicit
relationships between them. This scheme was inherited from the early social networking
8
sites like Friendster. The structure is based on a very simplified model of how people
interact offline. They explore the network by visiting each node and traveling on the
connected edges. As we will see later, these simplified offline notions of friendships
translate very poorly to the online world: services spend a lot of development time fixing
inconsistencies in the interface that result from this faulty transposition. The builders of
these networks have struggled to move away from this design for the same reason that
users still demand it: people are used to it. A scheme that accurately depicts the status of
user’s online relationships would allow services to provide a more relevant and
meaningful experience. Feld [1] and Granovetter [2] provide a good basic model for how
general offline social networks are organized and grow. Granovetter establishes the
semantics of talking about these networks, while Feld develops a complementary theory
to explain how they evolve. We will use their definitions of social ties to define a new
underlying structure for social networks.
In fact, explicit connections are replaced by activity dependant links. Granovetter
posits that looking at the small-scale interactions between people can give you a lot more
insight into the macroscopic behavior of the entire social network. He particularly uses
his small-scale definitions to derive rules on how information flows through a social
network. There are three types of ties that relate people: strong, weak, and local bridges.
This categorization is a direct mapping of how much work has been put into the given
relationship: the strength of a tie is characterized by the “amount of time, emotional
intensity, intimacy and reciprocal services” that are invested in them. A local bridge is a
tie that provides the only link between two sub-networks of friends. The links, as defined,
still provide the basic network properties one would expect. Granovetter discusses some
9
basic dynamics of these ties that allow us to predict their evolution. Relationships tend to
be transitive: if A is strongly tied to B and C, then B and C are likely to be very tied. As
the frequency of positive interaction grows, so does the strength of the tie. The stronger
the tie between people the more likely they are to be similar.
Also, Granovetter’s activity-based definition of links encompasses the notion of
potential relationships. This notion is very valuable to site builders since they would like
to promote a highly connected network. Instead of manually accumulating superfluous
friend relationships to access new parts of the network, users already have latent
connections to every other node. In fact, Granovetter very briefly discusses the idea of an
“absent” tie; this is a potential relationship. His paper, written in 1973, doesn’t deal with
these much since their evolution contributes almost nothing to the changes in the social
networks. In the offline world these are less important since the expected number of
conversions of potential relationships into a real tie is very small. However, in the online
world, the pool of available potential relationships as well as their conversion probability
is much higher. Web applications allow a user to quickly search through millions of
people’s profiles and contact them; converting the potential into a real relationship is
greatly facilitated. In fact, many social networking sites deem these relationships so
important that they compute and display an approximation of their total for each user.
The radius of a user’s social horizon is a measure of his potential connectedness, and a
barometer for the web site’s popularity.
A more realistic social network can be formed by expanding the types of elements
a node can represent. In the “Focused Organization of Social Ties”, Feld explains how
individuals tend to group themselves around foci of interest: these can be “social,
10
psychological, logical, physical entities”. In fact, browsing through the network isn’t
about moving between people, but around foci. The expansion of nodes around foci leads
us to a more powerful network representation. Indeed, Focus theory explains many of the
behaviors and dynamics that Granovetter discusses. For example, if A and B as well as A
and C are strongly tied it is because they are grouped around many foci. It is therefore not
inconceivable that B and C would also share a subset of those foci. The transitive nature
of ties also comes out of Focus Theory. The evolution of a person’s social ties is a main
part of Feld’s paper. Focus theory lends itself very well to online social networks since
they are laid out in a very interest driven interface. When looking at a person’s profile,
you get a comprehensive view of his foci: what organizations he belongs to, what people
he knows, and what his interests are. In fact, even if you don’t have an explicit
connection with the person, you are sometimes told how you overlap: how many friends
are in common, through whom you do you know them, which organizations do you
share? There are two forces that work together to drive the strengthening of old ties, and
the creation of new ties in these networks. The first is the pressure individuals feel to
associate themselves in stronger relationships: Feld notes that given a weak link, people
will try to find more foci that they have in common. The second force is the engineering
of the social space done by the builders of the site: keeping users active is a central goal.
It is not by accident that profile pages invite you to find common interests and
connections with people in the network. The stronger and more numerous the
connections a given user has, the more content and communication he will be subjected
to which will keep him active on the site. Any new piece of content or message now has
the potential to reach anyone in a user’s dense network of friends.
11
2.2. Signaling Theory
Signaling theory is a framework usually used for analyzing communication and
behaviors in the animal world. It proposes that agents have qualities which they wish to
signal to others. Judith Donath has proposed the application of this framework to the
online world. We will present the basics of Signaling Theory [3] and show that it lends
itself very well to the analysis of online social networking sites. In fact, it will allow us to
precisely determine the trade-offs of design decisions for problems at any scale.
At the base of the theory is the idea that each agent has qualities that they wish to
communicate. If the message is sent to the intended receiver we call the message a signal,
while if it is overheard than it is called evidence. The goal of this transmission is to
somehow alter the receiver’s beliefs or behaviors. However, determining whether a signal
is reliable can be problematic: we live in a world where deception can have many
benefits. Signals have costs and are reliable when “they are beneficial to produce
truthfully, and prohibitively costly to produce falsely.” An agent can advertise a quality
that exists in which case we have an honest signal, or he can be falsely promoting a
quality he doesn’t possess; this is a deceptive signal. An agent receiving the signal has to
use his aggregate experience to determine whether the sender was being or honest or
deceptive. The receiver will classify the signal as reliable if he can deduce that it was
always honest. The signal will be unreliable if there are too many deceptive
communications. In fact, the abuse of deceptive signalers destroys the reliability of
honest ones: if a signal is too easy to produce deceptively than it can’t be very reliable.
12
The application of signaling theory as an analysis framework will allow us to find
flaws in current social networking interfaces and design better ones. Signaling theory is
very well suited for analyzing the small-scale features of online social networks because
each action can be broken down into measurable costs and benefits. In this case, the
agents are the users of the service. The cost of using a feature can be calculated by
estimating the number of clicks needed to finish a task, or the time required to complete a
process. The builders must determine how expensive an interface should be based on the
benefits it provides: costs should be proportional to the benefits of signaling the quality.
Solving this economics problem will result in a feature that performs as desired. For
example, the design of a registration process on a college social networking site can
determine how honestly a user signals his membership to his school. Choosing to be a
part of a school’s community has many benefits: the site’s feature set allows each
member to have a profound local impact. If the users are allowed to pick their school
from a list without verification, then being part of the school’s online community will be
a poor signal that the user attends the college. In fact, there is almost no cost in picking
the wrong school out of the list and falsely registering for another university. In order to
provide a more honest signal to its users, the builders could require that new users use
their school supplied email address and click through on a verification email. The
additional work for real users is minimal: they need to take a minute to check their emails
and click on a link. On the other hand, a malicious user would have to do a lot of extra
work in order to create a dishonest signal: he would have to somehow find or steal a
working email account from that school and try not to get caught.
13
Finally, Signaling Theory is a useful tool in discussing the high level effectiveness of
a service. The builders of websites have tremendous power in creating a service that
matches their needs, often despite the wishes of their user base. However, the balance of
power is not one-sided because, in the end, the users have the right to choose which
service they want to belong to. Signaling theory provides the means to discuss the
competition between different websites in evolutionary terms: it is literally a game of
survival of the fittest. Each new service is competing against larger existing ones and
trying to sway the same user base. For example, the site builders of a fictional service
might want to attract new users by promoting themselves as the leading social network
for doctors. In this case, the signaling agent would be the entire service itself and the
quality would be the number of doctors using it. A press release with the size of the
service’s user base would serve to alter a potential member’s belief in the value of joining
the new site. Also, just like the loud roar of a lion, such a signal could also be intended to
intimidate competitors. Here the site builders could decide to risk their credibility by
weighing the benefit of providing exaggerated numbers versus the cost of an inquisitive
user or competitor catching them. In fact, sometimes site builders themselves may choose
to produce dishonest signals.
We have seen that Signaling Theory allows us to evaluate design decisions at many
different levels. The agents, qualities, costs, and benefits can be clearly defined in all
cases allowing users and site builders can make informed choices.
3. Signaling in Social Networks
3.1. User Level Signaling
14
Users will use the allotted feature set of a service to signal qualities that relate to the
perception of their online identity [3]. The signal sender and recipients are users of the
site. The actual signal will often be the output of a feature on the site: it can be anything
from a funny posting on friend’s message board, to a colorful profile customization.
There are many qualities that a user may want to signal in a typical social networking
site. For example, popularity can be gauged from the number of links to other people in
the network, the number of people who have left comments on their profile, or the
number of groups they belong to. Wit, humor and creativity are also important as most
sites allow you to personalize your name, pick wacky interests. Joining groups often
serve as badges of honor as they appear prominently on profile pages. Finally, technical
expertise is also commonly a signaled quality: users can gain notoriety for their ability to
cheat the system by adding video to the page when you’re not allowed to for example.
3.2. Service Level Signaling
The site builders achieve their goals by promoting the qualities of their service
internally to their users and externally to their advertisers, competitors, and investors.
Online services are an ecosystem of signals where the main challenge is finding the
balance between these constraints.
The site builders have multiple ways of targeting their current users. The first is to
become an actual member of the site and use User Level signaling to communicate the
advantages of the service. For example, employees could jump-start a service by
registering and completely filling out their profiles: their profile page would be an honest
signal that there is at least some activity on the site. Also, the design of the features on the
15
site, and the resulting user experience, can be engineered to promote a quality.
Throughout a user’s lifetime on the service, he is forming subjective opinions about his
experience; many of those belief changes are engineered by signaling. For example, site
builders might want to communicate qualities of security and privacy by prominently
displaying a privacy notice on the front page with links to profile viewing permissions.
Through the design of the features, site builders can signal many qualities to their current
users: fairness, transparency, reliability, and relevance. However, there is no stipulation
that the signals have to be honest; there are sometimes clear benefits to cheating.
There are many qualities that site builders try to communicate to receivers outside the
confines of their site. The signaling entity is the site builders and the actual signals are the
various characteristics of the online service. These characteristics are packaged together
to promote a quality very differently depending on the intended receiver. For example,
the size and demographic distribution of the user base are basic characteristics that can be
used for signaling. The site builders could issue a press release lauding important
registration milestones in a particular age group. This communication is intended to lure
potential users that match the thriving demographic to join the site: they will participate
in an active and dynamic community that is seemingly tailored to their needs. Also, the
communication might be a signal to competitors thinking of encroaching on their market:
the new milestones make the service unassailable. In fact, site builders might want to
signal the following qualities to receivers:
16
-
Potential Users: The site builders want to show that the network is large, dense,
relevant, and with limitless potential connections. The network will provide a
meaningful experience.
-
Competitors: The service has tremendous growth and imminent dominance in
the space. There is no need to compete with the service because it will be a
waste of energy.
-
Investors: The management team is smart, can make good decisions, and knows
how to monetize their service.
-
Advertisers: The service is a large site with a lot of consumers in an attractive
demographic.
3.3. Site Navigation Interfaces
Social networks set themselves apart from other online services by exposing the
connections between people and using them to navigate the site. We have seen that
members use these connections to shape their online identity while builders use them as
tools to promote their interests to a larger audience. These goals are often contradictory
and lead to inconsistencies in the navigation interfaces.
The site builders engineer their service so that it exhibits the qualities that they wish
to communicate. The success of a site is measured by the size of its network and its level
of activity. Features are designed to maximize the number of connections between users
to create a dense a network where information flows effortlessly. Therefore,
accumulating friend relationships is designed to be extremely attractive to the user: the
cost of making a new connection is very small, and the benefit of the new link is
17
immense. In her paper "Friends, Friendsters, and MySpace Top 8: Writing Community
Into Being on Social Network Sites." [6], danah boyd lists 13 reasons users add friends in
social networking sites. They were compiled from personal interviews she conducted
between 2003 and 2006 and are reproduced verbatim below:
1.
Actual friends
2.
Acquaintances, family members, colleagues
3.
It would be socially inappropriate to say no because you know them
4.
Having lots of Friends makes you look popular
5.
It’s a way of indicating that you are a fan (of that person, band, product, etc.)
6.
Your list of Friends reveals who you are
7.
Their Profile is cool so being Friends makes you look cool
8.
Collecting Friends lets you see more people (Friendster)
9.
It’s the only way to see a private Profile (MySpace)
10. Being Friends lets you see someone’s bulletins and their Friends-only blog posts (MySpace)
11. You want them to see your bulletins, private Profile, private blog (MySpace)
12. You can use your Friends list to find someone later
13. It’s easier to say yes than no
Not surprisingly, almost half the reasons (4,8,9,10,11, and12) are direct incentives put in
place by the site builders. Reason 13 is evidence that the costs in the interfaces are
purposely designed to push users to add new friends.
Connections between users are an unreliable signal for friendship. The imbalance in
the cost structure leads to the rampant accumulation of friend links, creating many
18
interface problems. We will discuss these in further detail in our case study of Sconex.
For example, browsing the site is challenging when each user has thousands of
connections. Also, it becomes very difficult to find meaning in the information carried by
the links. In the offline world, there is a limit in time and energy to maintaining
friendships. I can call up my 10 best friends and learn what they have been up to. In the
online world, I can easily accumulate a high number of relationships at no maintenance
cost: how can I sift through updates for 5000 friends? Often times, we see the more
advanced web sites trying to fix the inadequacies stated above. Since the reasoning
doesn’t come from a comprehensive analysis of their problems, their solutions are
ineffective fixes. For example, Facebook.com relies on geographic portioning and status
updates, while MySpace.com uses an “n favorite friend” subset.
A naïve solution to the problem is to make adding friends an expensive process. It’s
possible to design interfaces with high costs, but a sufficiently patient user will almost
always be able to circumvent them. If builders keep raising the cost of performing an
action without increasing the benefit, the feature will eventually not get used. The users
will stop coming to the site if the interface is prohibitively cumbersome. Every feature
should be as clean, fast, and intuitive as possible to reduce the users’ frustration. A
solution will require designing better site-wide interfaces so that there is no incentive to
abuse the features.
In fact, we will examine a new interface for site navigation through Medina, a social
networking site based on the idea of exchanging knowledge. The design will use a more
realistic model of how people interact online by using the underlying network structure
discussed in Chapter 2. Also, activity weighted links will allow users to meaningfully
19
browse the network while still expressing themselves through their set of friend
connections.
4. Sconex: signaling from a builder’s perspective
Social networks are not constructed in a vacuum: the builders of these networks have
many design constraints aside from keeping their users happy. We will examine the
design challenges of building a social networking site through the case study of Sconex
(http://www.sconex.com). The internal and external constraints on services force builders
to make bad decisions and mistakes. Through the analysis of Sconex we will show that
some of the more serious mistakes can be avoided by using a comprehensive analysis of
the interface design using Signaling Theory.
4.1. History
Josh Schanker and I founded a company called Sconex in August 2004. The name
Sconex was an abstraction from the words “Stay Connected” and encapsulated our wish
to build a social networking based service to enhance offline relationships. We started
building a variety of web services aimed at different markets but had very little success:
we focused primarily on the 25-45 age group with products ranging from business social
networks to a site promoting “flash reunions” of user defined groups. After a
conversation with my 16-year-old sister Meya Laraqui, I quickly realized that high school
students didn’t have a safe service that was dedicated to them. Although she knew of the
largest social networking sites out there they offered very little that appealed to her
friends: they preferred to communicate over instant messenger, and email. In fact, at this
20
time in January 2005, the social networking phenomenon was still in relative infancy:
there were a small amount of large players who dominated their respective markets, but
small niche oriented social networks were still forming. Facebook was firmly entrenched
as the college social networking site. Business professionals used LinkedIn. MySpace
was replacing the ailing Friendster as the dominant social network for adults 18 and over.
Of these behemoths, there were no sites dedicated to providing a safe place for 14-18
year old high school students to hang out online.
We quickly built the first version of the site, and preloaded it with all the private and
public high schools that were indexed during the last government census. The service was
based around the high school themselves: during the registration process, a user picks his
high school and has to answer security questions in order to be granted access. Each high
school had a profile page and was the center of the user’s world: he could see the classes
offered, ask homework questions, and contribute on the school’s message board. We also
built a suite of traditional social networking features to make the service more
compelling: users had their own photo album, blog, profile, mailbox, and interest
sections. As we launched we were faced with our first challenge: we had no users. Social
networks are compelling because they can bring together a large number of people and
facilitate information flow between them. However, without a single node in the network
it is difficult to generate any value for potential users. We seeded communities on Sconex
by spending our social capital: I asked my sister to sign up as many of her friends as
possible, and to get them to sign up their friends. We recruited a fleet of college freshman
in the Boston area and gave them $50 if they could sign up 50 people from their high
school. This was relatively effective in bringing initial users to the site but didn’t make
21
for a compelling service: users were isolated in small unexciting communities and had
very little loyalty to Sconex since they were begged by their friends to join.
Initially, we had designed in a high cost to joining the service because we were very
concerned with keeping the school communities safe and preventing fraud. In order to
join Sconex, you had to be explicitly invited by a current member and you had to answer
security questions about your school in order to complete the registration. In terms of
Signal Theory, being part of a high school on Sconex was an honest signal that you
attended that establishment since the registration cost was so high. However, the growth
of our user base was still linear. We thus decided to eliminate the need to be invited, and
opened up registration off the homepage to get more growth. We realized that a formal
invitation from one student to another carried a lot of social risk: the inviter was vouching
that Sconex was interesting and leaving himself open to judgment from the invitee.
Students preferred to casually mention Sconex to friends and let them go to the site on
their own and register. As a result, Sconex had a strong nucleus of around 5 thousand
members, a good portion of which were returning regularly to the site.
The next challenge was to find a way to increase Sconex’s user base by a factor of
100. We devised a plan based on the characteristics of the small communities that we had
seen form on the site. High Schools range in size from less than a hundred students for
some private schools, to thousands for large public schools. Our initial approach to
getting users had randomly sprinkled a handful of students in many of the high schools
across the country. The security settings that we had implemented also meant that a user
couldn’t see the profile of a student unless they went to the same high school, or if they
were friends. This meant that we had a lot of isolated communities with very few users in
22
them. However, the reality of a community space like New York is that there are a lot of
cross connections between different high schools. If schools are close geographically
many students will know each other through various in and out of school activities. We
first decided to add more security permissions so that student’s profiles could be seen by
more people: friend’s of friends, and classmates of friends could now see your profile by
default. This broadened the horizon of possible nodes a given student could visit in the
network. Also, we created a new class of users called Sconex Reps who were special
ambassadors within their school. In fact, in a service like Facebook, each community is
large enough that they can stand on their own: if the feature design isolates small colleges
with 5,000 students, the users can still generate enough content to make their experience
compelling. On Sconex, we had to create features that allowed high schools to be more
visible to each other so that 10 1,000-student high schools could be the basis of a
meaningful experience. A pattern is starting to emerge of meeting unforeseen challenges
by redesigning elements on the site in an ad-hoc manner.
The growth plan was to concentrate on a small geographic area and try to grow
schools one at a time, hoping that growth would spillover into neighboring communities.
We focused on New York City first since it had a very high density of high schools.
When a new school was created, the first user would provide security questions for it. We
would then manually verify that the questions were valid by cross-referencing school
answers using publicly available data. If the school was in an area of interest, we would
contact the user through the system and motivate him to invite friends. When the school
reached a non-negligible size (around 50 students), we would try to elect a Sconex Rep.
The Sconex Rep was a coveted title because it meant that the user was in charge of his
23
school’s community. Reps had a special set of features at their disposal to manage their
school: they could edit teacher and class lists, create better security questions, and report
students that didn’t belong to the school. The Sconex Rep program was a real success and
there were usually competing candidates for the position. Sconex Reps had an outlined 5
step program to grow their school: fix the security questions, invite 20 friends, create
school groups, add teachers and classes, and ask school group leaders to join Sconex and
manage their community (ex: the Softball team).
As a result Sconex explodes in New York City, where almost 75% of all high
school students are using the service by the end of 2005. Sconex is also the most “sticky”
web site in the country: Sconex users spend more time and consume more pages on the
site then any other social network. In fact, the most active user base is the urban youth in
public schools in and around the city. As people connect with friends and family in other
states, very strong communities start appearing in Miami and Atlanta and a presence
across the country develops. At its apogee, Sconex is delivering 300 million page views a
month to millions of users and is profitable without taking in money.
In 2006, interest in social networking sites reached feverish levels. Facebook had
registered every college student in the US and was expanding into High Schools.
MySpace slowly started realizing that an ever-growing percentage of their users were
under the age of 18: they dropped their age limit from 18 to 16, and then from 16 to 13
(the youngest allowable by law). In fact, most of MySpace’s growth came from the under
18 demographic. Many high school social networking sites start appearing like
MyYearbook.com, Tagged.com, and ClassFace.com to name a few. Faced with a difficult
battle we joined forces in March 2006 with Alloy, a youth media and marketing company
24
based in New York. Since then, the intense battle to wrestle users away from competitors
has only intensified.
4.2. Money Shapes Design
One of the greatest drivers of interface and feature design in online social networking
sites is profit. In fact, much of the feature design on the site was geared towards honestly
signaling usage patterns to advertisers so that profits could be maximized and the service
could stay afloat. It was important to honestly communicate our traffic patterns since
there exists industry trusted monitoring tools that our signals would be verified against.
We will follow the evolution of how design priorities were modified to generate new
signals.
There are several important metrics that a service is judged on: the number of users,
the number of page views it serves, the unique number of users who visit in a month, and
the average number of pages each user sees. For most social networking sites the revenue
model is tied to advertising based on usage. There are only a handful of successful niche
social networks that are able to charge a premium fee to use some part of their feature set.
For example, Classmates.com allows users to create profiles and search for their
classmates, but charges to get their contact information. In fact, the problem of finding
new revenue streams in large social networks like MySpace or Facebook is not solved.
The path that Sconex traveled is fairly typical of social networking sites and we will
discuss it further.
When Sconex first started getting traction and generating a non-negligible amount of
page views it became useful to invest time into monetizing that traffic. The Google Ads
25
program was the obvious first choice: there were no minimum impression thresholds to
be a part of the service, registration and installation were straightforward, and Google
sent a check within 30 days. In online advertising, the term CPM (Cost per mille) is used
to measure the amount of revenue generated by one ad during a thousand impressions.
Google Ads was easy to set up but had a very poor CPM, which means that each page
impression was poorly monetized. Under this scheme, the main design goal of the site
was to extract as many page views per person as possible. Since acquiring a new user
takes a lot of work, the more revenue that can be derived from him the better. Also, since
the largest financial burden on the organization was paying for servers, revenue would
scale at the same rate as costs. Increasing the CPM and number of page views, and
decreasing the cost to serve one page maximized profit.
As Sconex got larger, we started developing relationships with more advertising
networks. They could offer us better CPMs because they were reselling our ad space to
targeted buyers. However, our previous emphasis on maximizing page views was now
seen as a fault: advertisers are interested in getting a large reach, not pummeling the same
user with the same ad 100 times. In order to get to the next level of revenue goals, and
maintain the organization afloat, we had to shift our focus onto features that would attract
more unique users. During this period we added public contests and school polls; these
features were fun for registered users but also gave unregistered users content to look at.
The hope was that unregistered users would stumble onto the site and look at contests,
and the ads next to them, thus increasing Sconex’s number of unique users per month. If
even a small percentage converted to registered users, we would have also increased our
user base.
26
As part of Alloy, we had access to many direct relationships with very large
customers. The goals in feature design shifted again: with the potential of more lucrative
deals, we would build features that tied directly into custom promotions we were running.
Often these new promotions would be at odds with our current user base. The most
striking example is a model search contest Sconex hosted for Delia’s
(http://www.delias.com), a clothing company geared towards white suburban teenage
girls. Although the contest was a success, there was a tremendous amount of friction
between the predominantly white contestants and the regular Sconex users.
It seems that social networking sites have their priorities reversed: it is ironic that the
users that make up their service are often the voices that get drowned out. The builders of
social networking sites have many qualities that they wish to communicate; unfortunately
their users are a minority in the audience. Creating a valuable experience for users often
comes second to creating value for advertisers and investors. As we have seen, although
users are mostly interested in sending signals to other users, site builders are sending
signals to their users, advertisers, customers, parent company, investors, etc.
4.3. Stimulating Information Flow and Visibility
Increasing Number of users
There are several interfaces that we built at Sconex in order to try to increase the
number of users on the site. The first was to make the accumulation of friend
relationships a sign of popularity. Each person’s number of friends was prominently
displayed on their profile. In most social networks, a user’s popularity is often measured
by the number of friend connections they have amassed. For Sconex, being in the socially
27
unforgiving world of high school, this effect was all the more prevalent. Also, if a person
had a small number of connections, belittling labels would accompany the number. For
example, on a profile page it would say “You only have 4 friends”. These tactics were
very effective in getting people motivated to invite more people from their school to join.
The spirit of competition and school pride were also motivating factors. Each school
had its own profile page with the number and breakdown of students prominently
displayed. For many schools we were able to an estimate the number of students
attending from government census information. This allowed us to display the percentage
of students using Sconex and served as a goal the students could track as they tried to get
their school to 100%. Although creating a profile page for schools in a high school social
networking site seems obvious, Sconex was one of the few to do it. For example,
Facebook only started created school profiles in early 2007, more than 2 years after the
site launched. As we have seen, the user does not dictate feature design: Facebook
probably didn’t feel it needed to aggregate school data for its users to meet its goals.
Also, we developed a “Nearby Schools” feature that appeared as a module on school
profiles. This was meant to spur school rivalry, broaden the number of users visible
within a couple clicks, and form super-communities of schools. School profiles drove
students to invite more of their friends.
With the success of previous popularity experiments, we tried to get users excited
about the number of people they invited. We kept track of when a new student registered
as a result of an invite link, or invite email from a current user. Each users profile page
had the number of users they were responsible for bringing to the site. Unfortunately, the
28
feature was unsuccessful: students didn’t like to be held accountable for the people they
brought to the site.
In order to get current users to invite their friends, we tried to provide a compelling
feature set. We found that aggregating connection data and effectively displaying it back
to the students could be positive drivers of growth. Also, linking up smaller communities
with their neighbors created a competitive landscape where students wanted to outgrow
their rivals.
Increasing Page Views Per Person
The major goal of the website is to increase the number of page views. For the users,
it means that the site has enough relevant content that they can consume a couple hundred
pages in one sitting. For the builders, it means that each user is consuming a couple
hundred ads each day. Therefore builders focus on providing a compelling feature set that
continuously generates new relevant content for the users. However, most social
networking sites have a lot of security constraints that restrict the flow of information
through the network. For example, on Sconex, if you have no friends in other schools you
can only access content from people in your school. Users can quickly get isolated into
small communities with very little activity. You can design a more compelling service by
maximizing the effect of any action in the community: even a little activity that is very
visible can provide a compelling experience.
At Sconex, we tried to institute a design philosophy that every action should have a
reaction. Any user input into the system shouldn’t just increase the size of our databases,
but should engage other users. We designed a suite of alerts that appeared on a person’s
29
homepage to notify them of any relevant activity. An alert would appear if they had a
friend request, a comment on their profile, a comment on their blog, or pending group
requests. These alerts were a motivation for users to come back to the site every day and
check what was new. Attaching emails notifications to alerts was a powerful tool in
getting users back to the site. However, we used this very sparingly for only the most
important notifications as to not lose credibility.
Another approach to increasing page views was to bubble up and aggregate content.
Many of the objects on the site had an associated profile so that small communities could
form around them. Not only did people, groups, and schools have profiles but so did
classes, teachers, and interests. Each object had a picture, membership information,
message boards and paths to other objects. For example, the teacher profiles had the lists
of classes the teacher taught, interests had a “Related Interest” section based on the most
commonly shared interests from the member list. There we also “Top Lists” for almost
every type of content. In communities where there wasn’t much data, using different
ways of visualizing the data gave the users more engaging content.
New User Experience
First impressions are always very important. When a new user joins the service there
is a good chance that he will be in a community that isn’t completely developed. He will
at the very least have very few connections in the network. Getting enough relevant
information to a weakly connected node in a sparse community was a challenge.
In order to create a better new user experience, a new user was automatically added
to a couple of groups, was sent a private message, and sometimes a member of the
30
Sconex Staff was added as a friend. The goal was to have the user see a vibrant
homepage when he finished registering: the alert module at the top of the page would
lead the user to his welcome message, a “Recent Group Postings” module showed new
content from the automatically joined groups. For example, about the first half million
users on Sconex were added to the Sconex Central group up registering. The shear
number of members in the group resulted in new many messages being posted every
second. A new user would have a starting point in exploring the site; he could see that
although his community was sparse, other parts of the network were very exciting.
4.4. Challenges from Design Decisions
We faced many challenges at all phases of Sconex’s life: we very often made design
decisions to solve short term problems knowing full well that we would have to deal with
the consequences down the road. However, when faced with the constant struggle of
staying afloat in a startup company, any progress is welcomed. In this section we will
discuss some of the challenges that we faced, while in the next we will analyze their
causes using Signaling Theory.
As Sconex became larger, site navigation became a major challenge. Our efforts to
create dense networks where a lot of information was being exchanged on each link were
successful. Users were inviting a lot of their friends to fill out their schools, but they were
also meeting a lot of new people on the site. The number of friends the average user had
was on the rise. In the most extreme cases, some users had up to 50,00 friends which was
represented a couple percent of the total user base. On MySpace this would be equivalent
to having a couple million friends on the site. The number was an issue in that the friends
31
feature was growing much faster then our scaling plan had anticipated. Also, many of the
features that we had designed around friends were losing their relevance and become an
infrastructure burden. For example, it is hard to retrieve and organize the most recent
blog posts of your 10,000 friends. We saw the same effect for groups and interests. In
fact, on each profile page we listed the name of the groups they had joined. Soon
retrieving and listing 10,000 group names was technically and visually cumbersome.
Traditionally, if you wanted to browse all the groups that a person was involved in you
would be sent to a “Browse Groups” page where 10 groups at a time would be displayed.
We had done studies to find that almost 90% of people don’t go past the first couple of
pages. It became clear that this interface was hugely ineffective as users only consulted a
small random sample of the total number of groups.
Site performance suffered at the expense of our extensive aggregation needs. As we
have seen, exposing a lot of interesting membership calculations and providing a lot of
data aggregation pages was instrumental in getting the Sconex community off the
ground. However, this content requires a lot of work from the site infrastructure in order
to keep it up to date and relevant. We had used aggregation as a crutch when Sconex was
small to simulate a denser network but we were having a hard time supporting it as it
scaled. The aggregation tools we had designed had also lost a lot of their relevancy. It
became more difficult to tease out the most relevant data: there were too many links
producing too much data, and no good way to pick which one was the most important.
For example, the Recent Group Posts module designed to help the new user experience
was getting swamped. Even if a user had only 500 groups, each one was producing a
couple hundred posts a day. We tried to design new interfaces by looking more generally
32
at most recently updated groups for a person, but there was again much more
information than one human could digest.
The management of the user base was also a constant struggle. The users of any
service are creatures of habit that will have a strong resistance to any changes. The worst
change is the removal or modification of a feature. Although we were excited that our
users felt very strongly about our service, it made it a challenge to fix interface or feature
issues we had created. We improved our communication strategy to get more input and
guidance from them, and leaned on the Sconex Reps to convince the rest of the users a
change was worthwhile.
Conversely, not giving in to the requests of your users can be very dangerous. A
common feature request was to allow profile customization. Although there was a
tremendous outcry for more personalization, we didn’t want to anger our advertising
networks by drowning out their ads. In fact, the personal profile page was the most
viewed page on the site and contributed a lot of revenue. Faced with our lack of
flexibility, Sconex users found very resourceful ways to personalize their experience:
they decorated their first and last names. For example, by adding ASCII characters users
would transform Lady Tinkerbell into *~*LAD!!*~*T!NK3RB3LL*~*. Uses would also
very frequently change their names to coincide with what was going on their lives. This
little bit of creativity created tremendous interface problems for us: you were no longer
able pick your friends’ names out of a drop down list, searching for users by name didn’t
work, and exotic ASCII characters sometimes confused the database and the browsers.
4.5. Signal Theory Analysis
33
Many of the challenges that Sconex faced when there were millions of users were
based on decisions made when there were only 5 thousand. We will see that many of
these problems came from designing a poor cost structure for signals on the site.
When trying to attract users, site builders try to design the most efficient interfaces
possible: the amount of energy a user expends to become part of the service should be
minimal. In a world without dishonesty and cheating, the cost to becoming part of Sconex
would ideally be zero: we would like every high school student in the world to have the
quality “Sconex Member”. In order to build realistic interfaces that admit users into the
Sconex community, the minimum amount of costs were added in order to keep the
signals for exclusivity and security as honest as possible. We have seen in the Sconex
history, that originally we had placed too heavy a cost on membership: a user needed to
find answers to security questions and get a current user to invite him. By eliminating the
second constraint we effectively lowered the membership cost of a user. We saw the
anticipated effect: more users were joining the site, but the Sconex community required
more policing to remove unwanted users. In the same vein, the cost structures associated
with honestly signaling friendship were purposely skewed. A friend request is an
extremely powerful tool for site builders because it generates an email that will bring an
inactive back to the site, and also signals activity in the system. We spent a lot of time
designing an extremely efficient friend request process in order to drop the costs to a
minimum. Again, it isn’t surprising that 9 months later we are faced with users with
much too many friend requests.
In fact, there was a disconnect between the assumed signaling costs between related
features. Interfaces lost coherency since the expectation and reality of the signal costs
34
were not aligned. While designing the “get a new friend” interface, we wanted there to be
a very low energy cost in completing the process. The service would gain many more
members, and reinvigorate inactive users. On the other hand, when we were displaying
friend relationships in the service we were assuming that there was a much higher
acquisition cost. In these instances, we assumed that signaling online friendship was just
as expensive as signaling offline friendship. For example, the interface that allowed you
to see the latest blog posts from your friends made sense in this context. We were
assuming that you would be checking up on a small number of friends since each
relationship was so expensive to acquire and maintain. The same quality was signing
different cost contracts with different features. The problems stemming from irregular
cost assignments were exacerbated by the patching of broken interfaces. It is very
difficult to completely re-implement a broken feature without alienating large portions of
the user base.
As a site builder, we also had a lot of power that allowed us to cheat the cost
structures that we had implemented. By not playing by our own rules, we also created
more inequalities in the signaling structures. Site builders have complete control over
every aspect of their service: they have access to the data and the tools to modify their
service as they wish. With access to the databases and scripts to quickly do mass changes,
there is no signal that can’t be dishonestly produced with near zero cost. We had
complete control of every aspect of the system: we could send mass emails to users,
modify the database to create artificial friend or group relationships, etc. For example, by
shortcutting the group joining process and automatically adding new member to Sconex
Central we cheapened the quality of group membership. Users quickly learned that being
35
a part of a group has little meaning and we thus contributed to the frenzy of adding
groups as a means of self-expression.
Finally, the service as whole measured up poorly to competitors because of mistakes
in choosing which qualities to signal. For example, the Sconex users adamantly requested
profile customizations as a feature. However, we decided that building that feature would
make us too similar to MySpace, which had built its reputation on being a free-form
means of expression for teens. We didn’t want to duplicate their signals to prospective
users: by having a network orders of magnitude smaller, we would clearly be the loser in
the head to head match-up. However, it turns out the teenage demographic cared very
strongly about this feature. Ultimately, Sconex’s lack of ability to provide attractive
enough signals to compete with other larger sites has marginalized the service.
5. Medina: exploring activity based interfaces
In the previous sections, we’ve outlined the need for a strong unified approach to the
interface design of a social networking service. We have demonstrated that a simplistic
model of social interactions, and the conflicting needs of users and builders can result in
awkward navigation interfaces. Medina is a project that explores design decisions that
can provide meaningful site navigation while meeting the goals of all parties.
5.1. A marketplace for knowledge
The particular implementation example we have chosen to pursue is a social
networking site based around the idea of exchanging knowledge. The service is meant for
36
an audience like the MIT Media Lab. The lab is composed of professors, graduate and
undergraduate students, administrators, alumni, and sponsors. Each user has a very
specific set of knowledge expertise covering a potentially very wide range of topics.
Indeed, the research projects coming out of the lab are very different: projects range from
technology initiatives in developing countries to building quantum computers. The
people connected to the lab form a very rich network of heterogeneous knowledge.
However, in the offline world, access to this knowledge is difficult. There is first a
cataloguing and aggregation problem: it is difficult for a researcher to know every exotic
area of expertise that someone he is peripherally acquainted to might possess. Also,
physical access to the knowledge isn’t trivial. If a professor is the leading expert in his
field, he will most likely posses the answer to a related question; but it might not be
worth his time to spend his entire day fielding questions from inquisitive undergraduate
students. Finally, there is no permanent record of the knowledge transfers so that other
may benefit from past exchanges. The social networking site we propose to design will
aim to solve these problems among others. The word Medina, literally meaning "city" in
Arabic, is commonly used to describe the original historic part of a Moroccan city. These
old historic centers are still extremely active marketplaces. Analogously, the goal of the
Medina service is to provide an effective marketplace for knowledge. The service’s
interfaces will try to align the interests of the users and site builders by examining new
ways of connecting people and ideas. Also, we will propose a new paradigm for browsing
through the network.
Typically, a user of the system will alternate between two modes: searching for an
answer and providing one. When the user is searching for an answer he is relying on
37
someone in the network to expend energy to help him solve a problem. When providing
the answer he is the one expending energy. The user would like to get responses to his
questions very quickly, and usually not get bothered about answering questions unless he
gets something in return. In the offline world, people will usually exchange social capital
with their acquaintances in order to get/provide answers. Otherwise, there is some hidden
benefit: for example, a graduate student will answer the questions of his peers to establish
himself as the expert in a given subject. Users like to signal their expertise in a given
subject, but are rarely willing to provide help gratuitously. A challenge in the design of
Medina is honestly signaling the level of expertise of a user and his availability to
participate in a discussion. Also, Medina should allow users to signal the benefits of
being an active participant in the knowledge exchange.
Builders want to generate as much usage as possible. A large part of this task is
acquiring new users and having them visit the site on a regular basis. The builders are
thus forced to share a lot of the same concerns as its user base. Users should be able to
quickly track down an answer to their question, or find another member that will provide
a satisfactory answer with a high degree of probability. Members who are casual
participants shouldn’t be annoyed with requests even though they may be experts in their
field. As always, builders want to accurately signal that the service is a vibrant and
dynamic destination. In Medina’s case, this can be achieved by seeding the service with
organizational data already compiled by the Media Lab, as well as integrating external
data sources like archived mailings lists.
38
5.2. Activity-Based Interface
We have seen that the cost structures for signals in social networking sites are often
poorly designed. Firstly, there is competition between the qualities that the builders and
the users wish to signal. Secondly, there are many outside pressures that force the
builders to consciously make bad design decisions. Finally, features are often not
rigorously designed within a consistent framework. Just like any complex system,
features are added without taking into account previous assumptions. Services get trapped
by their own bad interfaces and can only apply superficial fixes instead of solving the real
underlying problems. The goal of the Medina project is to design a social networking site
where the basic elements of the service take into account the needs of all signaling
entities.
The first step to addressing our design goals is to define a new underlying structure
for the network. We are going to use Feld’s general construction of social networks
where all elements have equal importance. In fact, users, groups, interests, and ideas will
all be considered foci. This is a departure from how online social networks are
traditionally viewed: the network is a set of nodes representing people, and the edges are
the explicit connections they have defined. For example, Sconex was designed such that
each high school student was a node in the network and the user-supplied friend
connections where the set of edges. On Medina, both users and groups are considered to
be network foci: this view matches the intuition that a user can either find the answer to a
question directly in an interest group or by directly asking another member.
Signaling Theory will help us define how the links between foci should be created.
As we have seen with Sconex, the cost of manually creating links between nodes is
39
extremely cheap: adding a friend, or joining a group are no longer reliable signals for
friendship and group membership. In the offline world, it takes energy to create and
maintain a relationship so the resulting signals are harder to fake. Therefore, Medina
measures the activity between foci and uses the resulting score as an edge weight.
Therefore, the edges between foci represent the amount of energy put into the
relationships. Accordingly, if there is a lot of activity over time between foci then their
activity score increases; if the relationship is neglected the activity score should slowly
decay. Many of the design goals discussed can be achieved by measuring activity
between elements of the network. For example, users are able to quickly find the right
person to help them. A user might label himself an expert in “Social Visualizations” but
never participate in the group. The people who claim to know the most about something
aren’t always the most helpful. A person who is a learning novice but spends a lot of time
contributing to the group will have time to actually share his knowledge and end up being
more helpful. Therefore the member with the higher activity score will be the most apt to
help with a related question.
These activity scores also define a new means of browsing the website. Traditionally,
users will explore a social networking service by moving along the group membership
and friend links in the network. As we have seen, it is difficult to attach meaning to these
since they are so easy to create. On Medina, the connections between foci are defined by
the amount of activity between them: by providing a network visualization where links
with a high activity score are more visible, users can explore the network in a more
meaningful way. It is important to note that we are disassociating the idea of explicit user
supplied connections and edge links. For example, two users might send messages back
40
and forth to each other every day and thus have a very strong activity score between
them. They will appear very prominently in each other’s network. However, they may or
may not choose to explicitly label their relationship as a “Friend Connection”: explicit
connections are just another input to the activity score calculation.
Using this design, users can still express themselves through the public display of
their connections. Those who like accumulating links can coexist with users with a more
literal definition of a “Friend”. In both cases, the network visualization will provide the
same type of navigation functionality: the most meaningful connections for each will be
the most prominently displayed. From the builder’s perspective, the network will have
many opportunities for information to flow between users since there will be a high
conversion rate of absent ties to weak ties. In the case that a user’s numerous connections
might carry too much information, sorting by incident activity score organizes the data in
a relevant manner.
5.3. Site Functionality
5.3.1. Platform Specifications
Medina is a web application written in PHP version 5 and supported by open source
software. It is hosted on a server running Linux Fedora Core 6, using Apache HTTP
Server version 2.2.2 as a web server, and MySQL version 5.0 as a database server. The
service also relies on network visualization applications written for Flash 7 using
ActionScript 2.0.
5.3.2. Feature Set
41
The implementation of the Medina service is meant as a proof of concept of the
design outlined above. As such, we have tried to build a diverse feature set where each
important element is fully flushed out. However, we preferred to focus our efforts on
general interface design rather than providing tremendous depth and aesthetics to each
feature. The Medina service includes a suite of traditional social networking features.
Users must register to use the service and login with an email and password to gain
access. There is a registration page that only allows users with a unique “mit.edu” address
to join. The registration process is very simple and requires very limited information: first
name, last name, unique email address, and password. Once inside the site, the service is
made up of 5 main pages around which the major features are organized: the homepage,
the profile page, a mailbox, a blog, and the focus page. Most pages are organized in two
columns with sub-elements divided into modules. Let us describe the features in more
detail:
-
The homepage is a centralized information source for the user. If there are any
alerts from friend requests, unread mail messages, or unseen profile comments the
Alert Module appears at the top. The rest of the modules on the page give the user
a feeling for what’s going on the site: there is a module listing the most active
users on the site, and a module with the most active external mailing lists.
-
The mailbox is a listing of private messages that have been sent to the user.
Messages are either read or unread, and can be deleted. There is no notion of
threading, although message replies are tracked.
42
-
The blog is a time-sorted sequence of posts a user has made. Other users can
comment on each post.
-
The profile page is an aggregation of all the content for a user: as in Figure 1, it
contains sample modules for his friends, profile comments, basic information, and
questions he has successfully answered. Also, the page lists all the foci the user
has declared himself a member of.
-
The profile comments are publicly viewable messages that anyone can leave on
another user’s profile. A sample of the last comments appears on profile pages,
and a browse-page allows users to access all of them.
-
The focus page is another aggregation page. It contains a sample of the members
that have joined, the most recent questions asked about the focus, and the latest
associated archived mailing lists posts. In fact, when creating a focus a mailing
list can associated with the group: if the list is archived by the Media Lab its
message will be accessible from inside Medina.
-
The knowledge base of a focus is represented by a question and answer style
message board. Any user can post a question to a focus, and any user can provide
possible answers. The user who asked
-
The focus and profile pages have a network visualization that provides a new way
to access new interesting nodes.
-
A small set of administration tools are also provided. There are multiple user
levels that allow level specific security settings. In actuality, only two are used to
separate administrators from regular users. The main difference is that
43
administrators can change their display to a debug mode, and impersonate other
users.
Figure 1: Medina Person Profile Page
5.3.3. Measuring Activity
The structure of the underlying network has been changed to include people and
groups, referred together as foci, as the nodes in the network. The links between them are
edge-weighted links based on the amount of activity between the elements. The links are
unidirectional. Just as Feld and Granovetter postulated, the more effort is put into the link
between two foci, the stronger the tie between them. Conversely, if relationships are not
maintained they will slowly decay. The Medina service measures activity between foci
and updates the activity scores accordingly.
The traditional social networking features were included to provide a realistic means
of measuring activity between elements in the network. For each action a user performs
44
his activity score will be updated. The time cost and social importance of the different
possible actions were subjectively calculated and hardwired into the code. When an
action is performed its positive or negative cost is added to the activity score between the
elements. For example, viewing another user’s profile increases the activity score by one
point. Sending a personal message increases the activity score by 5 points; if the user
replies to the message then the score is increased by 10 points. In fact, viewing a profile
has very little cost, therefore the impact on the activity score should be very minimal.
When a user sends a message to another member, there are not only more clicks required
in producing the action but also the social significance is deeper as the user is initiating a
personal communication. If the recipient replies to the message then he is validating the
exchange as fruitful. The idea of decay in a relationship is implemented by docking the
activity score of each link by 1 at the end of each day. Also, the score between each foci
is archived so that the history of a link can be visualized.
The implementation of the scoring system matches up to our earlier definitions of
how the underlying network should be organized. Foci have activity scores with other
nodes in the interval [-∞;+ ∞]. Once two foci are connected the absent tie between them
is converted to a weak tie. One of the most common actions on a social networking site is
viewing the profile page of a focus (either a person or a group on Medina). Users will
casually visit hundreds of profile pages without investing much energy, past satisfying a
fleeting curiosity. One could ask what is the purpose of increasing the activity score for
such a mundane action? In fact, the addition of this small score will convert absent ties in
the network into weak ones. This will create dense connected networks, even for users
haven’t been members for a long time.
45
5.4. Activity Visualization
The activity scores between foci can be used to create a better browsing interface for
the service: the visualization can filter out weak ties and guide attention to the valuable
strong ties. Instead of providing long lists of connections, the interface can sort a person
or groups’ outgoing links based on activity. In fact, the elements with the highest score
will be the most relevant. This visualization will be prominently displayed on the person
and group profile page: it will be the main means of site navigation. For example, a user
has a question about using the WaterJet machine in the basement of the Media Lab. He
searches for the WaterJet group and lands on its profile page: he would like to find the
user that could offer him the most help. Traditionally, he would have had to randomly
look through the group member list to find a partner; perhaps he will increase his chances
of success and will look for an intersection with his close friends, or consult the list of
users who have labeled themselves as experts. These procedures don’t have a large
probability of getting our user a reliable response to his question. However, if the user
could quickly see a view of the most active users in the group, he could find a user with a
high probability of responding with a valid answer. In fact, these activity visualization
maps would be useful for user and group profile pages. On a user’s profile page, the
visualization would include the users and groups he had interfaced with. On the group
profile page, the visualization would have the users who were active. The visualization
has a zoom feature to see varying levels of network details. It also has a history feature
that animates the foci in the visualization based on the history of activity scores.
46
The actual visualization built for the Medina web site was a Flash 7 application
written entirely in Action Script 2.0. The application is embedded in the profile pages of
foci and passed both the focus type and its unique focus id. The flash application then
fetches a representation of the focus’ network through an XML interface. The data
collected includes information on the main focus and a listing of all the foci that are
connected to it. Each connected focus comes with its type, current activity score, and 30day activity score history. The activity scores were rescaled such that they represented a
real distance in pixels in the visualization and that all nodes would appear at the highest
zoom level. When designing the visualization, two layout options inspired by the email
visualizations of Viegas et al [5].were entertained. The first design is a circular
arrangement where the main focus is at the center of the visualization: the connected foci
are randomly placed on a circle with the scaled activity score as the radius. The foci
representing people are colored in red, and the foci representing interest groups are
colored green. In Figure 2, we are looking at an example visualization for the user
Jawad, who is most active with the users Carl and Peter, followed by the group “Sociable
Media”.
47
Figure 2: Circular View of Visualization.
A drawback of this visualization is that the second degree of freedom, the angle around
the circle, isn’t assigned a meaning. The second visualization option represents the foci in
a waterfall: the nodes with the highest activity score are closer to the top. For example, in
Figure 3, we have the visualization embedded on Jawad Laraqui’s profile page. The
interest group Sociable Media is the node with the highest activity score.
Figure 3: Waterfall View of Visualization Zoomed In, and Zoomed Out.
48
Ultimately, the waterfall layout was chosen because it didn’t contain the ambiguity of a
meaningless degree of freedom.
Though the two visualization options lay out the nodes differently, they have the
same functionality. First of all, they are a browsing interface: clicking on a node sends
the user to the profile for that focus. Also, the zoom controls in the upper left hand corner
allow the most active nodes to be visible at a high resolution. As the visualization is
zoomed out, a high-level view of a user’s activity connections emerge. For example in
the zoomed out visualization in Figure 3, the users seems to be casually browsing many
interests groups but not contributing much; this can be seen from the mass of green foci
at the bottom of the waterfall. This view represents a snapshot of a user’s contribution to
the service. Also, the application allows users to visualize the history of a foci’s
connections by using the movie controls near the bottom. The ensuing animation traces
the changes in activity scores for each node over the last 30 days.
5.5. Medina Evaluation
The Medina project was a proof-of-concept implementation of the activity-based
interfaces we have been discussing. As such, there are still open questions as to the
behavior of the interface under real world scenarios. The scoring of the features was
subjective and might not correspond well to the user’s expectations. It is expected that the
scoring coefficients will constantly be tweaked during the lifetime of such a service.
Also, the strength of an activity-based interface is that it removes absolute power from
the site builders to shape the network as they please. On the other hand, it also forces the
users to expose their usage patterns on the service. Users might not like the lack of
49
control over the browsing interface of their profile page, and feel frustrated by the
inability to easily fake their online identity. Finally, we are placing a burden on the users
who contribute a lot within an interest to be gurus for all others.
6. Conclusion
We have seen that poor interfaces in online social networking sites stem from the
inconsistent analysis of the evolving constraints placed on them. These constraints not
only originate from the users of the service but also the people that build them. In the
hopes of providing a better foundation for our analysis, we first revisited the underlying
structure of social networks. We argued that a more realistic network definition could be
utilized by using a more general definition for nodes, and using activity weighted links
between them. In fact, we saw that using activity was a more natural transposition of how
social network function offline.
Then we introduced Signaling Theory as a framework for analyzing social networks:
through the analysis of the costs and benefits, design decisions can be evaluated at all
scales. This provided the foundation for a consistent discussion of online interfaces. We
discussed how users signal qualities in order to shape their online identity, while site
builders try to resolve the constraints imposed by their members, advertisers, and
investors. We saw that current social networking interfaces fail to adequately address
both sets of goals and contain important inconsistencies.
Then this framework was used in the case study of Sconex, a high school social
networking site for teens. We saw the importance of building consistent interfaces that
50
satisfy the needs of all signaling entities. Also, we saw how bad design decisions in terms
of signaling can produce life-threatening interface issues.
Finally, we discussed a possible solution with the activity-based interfaces of Medina.
The underlying network structure discussed in Chapter 2 was used. A network
visualization tool emphasized the most active foci for a person or interest group allowing
for a more meaningful way to navigate the site. The users of the site were still able to
express themselves, while the site builders could satisfy the external constraints imposed
by the business side of the service.
7. Bibliography
1. Feld S L: ‘The focused organization of social ties’, American Journal of
Sociology, 86, No 5, pp 1015—1035 (March 1981).
2. Granovetter M S: ‘The strength of weak ties’, American Journal of Sociology, 78,
pp 1360—1380 (1973).
3. Donath, J: “Signals, cues and meaning”,
http://smg.media.mit.edu/classes/IdentitySignals06/SignalingDraft.pdf
4. Donath J, boyd d: “Public displays of connection”, BT Technology Journal Vol.
22 No 4 October 2004, pp 71-82.
5. Viegas, F et al.: “Digital Artifacts for Remembering and Storytelling: PostHistory
and Social Network Fragments”, HICSS 2004 (conference paper)
6. boyd, d: “Friends, Friendsters, and MySpace Top 8: Writing Community Into
Being on Social Network Sites.”, First Monday 11(12), December 2006,
http://www.firstmonday.org/issues/issue11_12/boyd/index.html
51